DE4102412A1 - Modulation mode recognition for baseband signal transformation - subjecting intermediate signal, or couples band signal to parameter estimation - Google Patents

Abstract

The modulation mode recognition methods has a modulated, complex base band signal, initially of unknown modulation mode, which is transformed into numerous, independent intermediate signals, forming a base for the determination of the modulation mode in an evaluation circuit. The individual intermediate signals, and/or the original complex base band signal, are subjected to parameter estimation of an autoregressive model. From the estimated parameters, and/or derived prediction errors the modulation mode is determined by intercomparison or comparison with preset threshold values. USE/ADVANTAGE - For signal transmission systems, with simple and effective layout, and high reliability.

Description

The invention relates to a method for modulation species recognition according to the preamble of claim 1 and an arrangement for carrying out the method. A such a method is already known from DE 34 03 187 A1 known.

Messages corresponding to a Mon Formulation type converted. To realize that news be sent according to a certain type of modulation or to get messages correct according to their modulation type it is necessary to demodulate the modulation type of Detect signals.  

The modulation type detection from z. B. digital messaging tensensalen mostly builds on an attempted demodulation on. If the attempt at demodulation succeeds, the Demodulation type recognized, otherwise the signal as not classifiable rejected.

In the method known from DE 34 03 187 A1 after a certain, initially still unknown modu Complex baseband signal modulated in meh Transfor independent intermediate signals mated, namely a nonlinear transformation of the Real and imaginary part of the complex baseband signal x performed to be the amount and phase of the signal to be able to count. The intermediate signals thus obtained bil the basis for determining the type of modulation in a downstream evaluation circuit.

Calculating the amount and phase from the complex Baseband signal x is usually very computationally intensive. It only interactive algorithms exist. These have ent neither linear convergence (such as that for the one in the Ar article by T. C. Chen: "Automatic Computation of Exponenti as, Logarithms, Ratios and Square Roots ", in" IBM J. Res. Develop. ", Vol. 16, No. 4, July 1972, pp. 380 to 388 written coordinate rotation digital computer (CORDIC) algorithm used) and therefore need at least as many arithmetic steps as the accuracy of the result nisses or quadratic convergence with the Un Support of tables that have to be filed.

It also enhances the nonlinear transformation of Real and imaginary part of the complex baseband signal x in  Amount and phase the noise. Does the amplitude of the Useful signal, the roughness becomes at small amplitudes predominate in the overall signal and accordingly the Determine phase. The phase evaluation then becomes full wrong phases are equal to phases of high quality tet. So it comes with low signal-to-noise-Ver SNR conditions tend to produce incorrect results.

The object of the invention is a method of to create the type mentioned that with a poss simple arrangement can be carried out and Reliability as high as possible with the modulation types has identifier.

The achievement of the object is in Pa Claim 1 described. In the other claims are advantageous training and further developments of the Invention according to the method and a preferred arrangement for Execution of the method according to the invention and its partial training and further education described.

The inventive method provides that the individual NEN intermediate signals and / or the complex baseband signal x each a parameter estimate of at least one autoregressive model and that from the estimated parameters and / or from these parameters derived prediction errors by comparison at least some of these parameters or prediction errors with each other and / or with predetermined threshold values the type of modulation is determined. (Autoregressive models are, for example, from the digital spectralana lysis already known (see e.g. Chapter 7 in the textbook  "Digital Spectral Analysis" by S.L. Marple Jr., pp 189 to 203 (Prentice-Hall, Englewood Cliffs, New Jersey, 1987)).

A major advantage of the method according to the invention consists in the fact that the respective type of modulation relatively easy and fast as well as with a relatively high one Level of reliability can be determined.

The method can be used to determine any modu lation types are used, especially for determination of analog or digital modulation types.

In a special embodiment of the invention The individual intermediate signals are expediently used in each case by a linear or non-linear Transformation from the complex baseband signal x er testifies, preferably the majority of the transformation nonlinear transformations of preferably never their order, preferably first, second, third or fourth order.

The individual transformations can be in time other and / or preferably carried out in parallel will.

For the individual intermediate signals and / or the complex Baseband signal x itself advantageously become authors gressive models of different orders, preferably lower order, preferably first, second or third ter order used to the special signals (intermediate or baseband signals) have been adjusted.  

The autoregressive models used preferably work on the basis of Levinson or Schur algorithms (the Levinson algorithm is described, for example, in section 7 of the above textbook by SL Marple Jr., the Schur algorithm, for example, in "Journal für pure and applied mathematics ", Vol. 147 (1917) pages 205 to 232), the prediction errors are expediently normalized to values 0 1 for faster calculation.

To detect digital types of modulation and / or harmonic signals, in a particularly advantageous embodiment of the method according to the invention, the complex baseband signal x is subjected directly to the parameter estimation of an autoregressive model of first and / or second order and / or first into the intermediate signals x ² and / or x ⁴ and / or | x | ² and / or (| x | ²-m _{| x | ²} ) and / or (| x | ²-m _{| x | ²} ) ². These intermediate signals are then each subjected to a parameter estimation of a third-order autoregressive model.

Preferably, the prediction errors α _d1 , α _d2 , α _s , α _f , α _c , α _m , α _t derived and preferably standardized from the individual parameter estimates are then at least partially related to one another in the evaluation circuit and these individual relationships

α _y / α _z (y, z ∈ {d1, d2, s, f, c, m, t}; y ≠ z)

individually compared with _{predetermined} threshold values S _yz , the present type of modulation being determined by multiple comparison of the differences resulting from these individual comparisons [(α _y / α _z ) -S _yz )].

In the following, the invention is exemplary based on the Er recognition or determination of digital modulation types hand of the figures explained in more detail. Show it:

Fig. 1 is a block diagram of a demodulator with a Modulationsartenerkenner,

Fig. 2 shows the block diagram of the demodulator of FIG. 1 for PSK2 or PSK4 zusätz with a handy device for carrier recovery and Baudratenabschätzung,

Fig. 3 is a table showing the Determi niertheit (D) or non-determinism (N) for various digital modulation types and for different non-linearities

Fig. 4 is a table showing an advantageous choice of relation to each other to be set Prädiktionsfehlerpaaren for determining the modulation type of digital modulation modes,

Fig. 5-6 advantageous embodiments of the arrangement according to the invention for carrying out the method according to the invention.

The demodulation device in Fig. 1 comprises a demodulator DM, which is intended to demodulate the (e.g. digitally) modulated complex baseband signal x and to output the demodulated message signal D (e.g. a digital data signal) at its output.

In parallel, the signal to be demodulated is a module tion type recognizer ME supplied, the signal x analy based and the present (e.g. digital) type of modulation, with which the signal x is modulated, and the one control signal S corresponding to the detected type of modulation to the demodulator DM, which thereupon accordingly the detected type of modulation demodulates the signal x and at its output the demodulated message signal D issues.

Such demodulation devices are used for for example in surveillance systems for HF receivers.

The demodulation device in FIG. 2 is provided for the digital modulation types PSK2 and PSK4 (PSK = phase shift keying) and differs from the demodulation device in FIG. 1 only in that an additional device for carrier recovery TR or for baud rate estimation BA is provided, which gives an additional control signal S 'relating to the carrier frequency and phase, baud rate and phase to the demodulator DM.

The Modulationsartenerkenner ME in FIG. 1 has to assign the task to the individual signals classes of modulation modes and to create a ranking of the gnale to be processed Si.

There are e.g. B. the following classes of modulation types differentiated:

• a) PSK (phase shift keying) + noise ("PSK2" or "PSK4")
• b) ASK (amplitude shift keying) + noise ("ASK2")
• c) FSK (Frequency Shift Keying), including MSK (Minimum shift keying) + noise ("FSK2", "MSK2")
• d) harmonic signal (sum of sinusoidal signals) + Noise ("sine")
• e) Noise.

Speech signals (radio) are not recorded. Here you can Noise and harmonic signal in about 15 ms against each other the alternate. In particular reflected in the past a fixed frequency grid and intoned speech the ASK class to the modulation type recognizer. By previous segmentation can, however, belong together rende signal parts summarized and so a "hiking" the Speech signals can be prevented by the channel grid.

If one leads digital modulation types over different Non-linearities become dependent on the modulation The data in the message may or may not be destroyed.

If the data are assumed to be random, they cannot be predicted. Once destroyed, it is a prediction possible and the signal is rather deterministic. If one removes the deterministic part of the signal, so there is usually only an unpredictable signal. If you check the degree of predictability using the nor prediction error of an autoregressive model after, the predictability before and after ver different nonlinearities to certain modulation ares ten be closed.  

The method of linear prediction used in autoregressive models is based on the autocorrelation function r _{k of} the signal x _k and consists of the solution of the system of equations

or short

R · ª = · e _o (1b)

where r _{n is} the n-order autocorrelation coefficient, r _n * the corresponding complex conjugate is efficient, a ₁ -a _{n are} the coefficients of the auto-regressive model used and α is the prediction error (a scalar). R is a (Hermitian) Töplitz matrix of the autocorrelation coefficients r _i , r _i * (i = 0, 1,... N), ª the vector of the coefficients 1 , a ₁ ... a _{n of} the autoregressive model used and e _o der zeroth base vector. The normalized prediction error is

α = / r _o (2)

A deterministic signal has the normalized prediction on error α = 0, white Gaussian noise in the prediction error α = 1.

For various nonlinearities and digital modulation types, the table in FIG. 3 for the determinacy (D) (α = 0) or unpredictability (N) (α = 1) can be derived.

In the table, x01 stands for a first-order linear predictor (i.e. no non-linearity) and x02 for a second-order linear predictor (also no non-linearity), while x ² , x ⁴ , (| x | ²-m _{| x |} ²) , (| x | ²-m _{| x |} ²) ² represent non-linearities with respect to the signal to be transformed x, where m _{| x |} ² is the arithmetic mean of | x | ² and | x | ² is the square of magnitude of x.

The linear predictors x01 and x02 are introduced to To be able to differentiate between MSK2 and FSK2, since FSK2 in the free quenzspektrum has two spectral cluster points, while MSK2 has only one spectral cluster point. A pre first order diktor can use the two spectral clustering do not replicate FSK2 points, one predictor second However, this can be done in order. That's why with FSK2 Transition from predictor first to a predictor second Order a characteristic increase for FSK2 predictability given.

The remaining linear predictors for those by the intermediate signals obtained from nonlinear transformations are advantageously third order.  

The classification of the digital modulation types is done by threshold comparison. One forms the quotient of two standardized prediction errors α, in which the relationships of determinacy D and non-determination N are reversed as far as possible or at least differ in one of the two measured values. The table shown in FIG. 4 gives an overview of a preferred selection of threshold value comparisons with which the respective type of modulation can be correctly determined with a particularly high degree of probability.

Here is when the numerical value at the cross point of two types of modulation is greater than a threshold that Modulation type of the column, if less than the threshold, to take the modulation type of the line.

This is a heuristic linking of the characteristic values. Of course it is also possible to evaluate the To leave characteristic values to a pattern recognition process, such as e.g. B. the linear and quadratic polynomial classifiers (see, for example: J. Schürmann: "Polynomial classifiers for the Character recognition - approach, adaptation, applications "(R. Oldenbourg Verlag, Munich, Vienna, 1977, pages 174 to 177)) or the neural networks (R. P. Lippmann: "Pattern Classification Using Neural Networks ", in IEEE Communica tions Magazine, November 1989, pages 47 to 64).

An n · n matrix A for linking the α values to decision functions can, for. B. can be obtained on the basis of a statistical evaluation and consists of a number of positive and / or negative numerical values A _ÿ (i, j = 1, 2,... N), here as an example for n = 7:

The Vek is used to classify the type of modulation gate

multiplies the above matrix A and the vector α and it holds the classification vector k

k = A α . (5)

The largest entry of the classification vector k indicates the type of modulation. In the example, the entries correspond in turn:

• 1. carrier (sine),
• 2. amplitude shift keying (ASK2),
• 3. Frequency shift keying (FSK2),
• 4. Minimum Shift Keying (MSK2),
• 5. Binary phase shift keying (PSK2),
• 6. Quartenary phase shift keying (PSK4),
• 7. Noise.

The modulation type detection according to the invention breaks down into this particular embodiment in two or three Parts: a) determination of characteristics, b) selection of the one Class-related signals and possibly c) rank determination within the desired class.

Another embodiment of the inventive method rens is explained in more detail below.

The types of modulation are based on a few Parameters that essentially reflect the power density spec trum (2nd order spectrum) and 4th and 8th order spectrum match, recognized. For the modulati considered Types are spectra of odd order in the expected value Zero.

The 4th order spectrum provides information like the signal from Gaussian noise is different.

The power density spectra are auto-regressive Approximate models. This has the advantage that a few para meters to describe complex power density spectra suffice. This fact can also be found in the Speech coding with low bit rate can be used.  

The crucial question is always what percentage of the energy in the signal low-order autoregressive model.

The algorithm consists of calculating the correlations values up to the specified low order for a data block. Levin already mentioned above For example, the algorithm is used once per block Determination of the by the autoregressive model not before predictable energy calculated. It provides a preference wise standardized prediction error α.

An example of deterministic signals is the sine, the itself as a real signal by a real autoregressive Fully predict second order model. The Prediction error α becomes 0. In the complex case it becomes for a sine wave only the first order full prediction needed.

The example of a stochastic signal is wide bandy Gaussian noise, whose energy is autoregressive model cannot be explained. Here results the prediction error α = 1.

An essential feature of the invention for the construction of the features is that the complex baseband signal x (k) transports data and is therefore not completely predictable, since otherwise there would be no interest in communication between the transmitter and receiver (k corresponds to the respective sampling times t _k ). If the signal x (k) is now subjected to a transformation T which eliminates the data dependency of the signal, the signal should be more predictable. Different transformations (non-linearities) can be found for the different types of modulation, in which the data is optimally deleted. The ratio of the parameter α is then examined before and after the transformation T

α (T (x (k))) / α (k)) <threshold.

This ratio is decided against a threshold.

It is characterized, for example, by the following features went:

• a) A real third-order autoregressive model is used for the mean-free square of the complex baseband signal x (prediction error α _m ). At PSK, the amount of the signal destroys the data. At ASK, the data dependency remains.
• b) A complex first order autoregressive model is applied to the complex baseband signal x. High predictability indicates a carrier or slow data transmission with good SNR. This prediction error is denoted by α _d1 . With a sine and SNR »1 it gives the prediction error of the unexplainable energy to 2 / SNR.
• c) A complex third-order autoregressive model is used for the complex signal square (characteristic value α _s ). With PSK2 the data is destroyed by squaring, with PSK4 this does not happen, with FSK / MSK only in exceptional cases.
• d) A complex third-order autoregressive model is used for the fourth power of the complex signal (prediction error α _f ). With PSK4 all data are now destroyed by this non-linearity.
  The classification features can be z. B. are formed as follows:
• e) If the square of the amount of the signal cannot be predicted beforehand, α _m threshold, it is concluded that noise or continuous types of modulation.
• f) If the ratio of the prediction error of the amount square of the signal to the prediction error of the signal is greater than a threshold α _m / α _d1 threshold → ASK, the conclusion is ASK (Morse).
• g) The ratio of the prediction error of the complex square of the signal to the prediction error of the signal results in a threshold decision between harmonic signals, FSK and MSK against PSK α _s / αd threshold → FSK, sine, MSK.
• h) A ratio of the prediction error of the complex squared signal to the prediction error of the fourth power of the complex baseband signal distinguishes between PSK2 and PSK4 α _f / α _s threshold → PSK2.

The classes are then derived from the characteristics e) -h) Thresholds formed.

According to a linear evaluation function, the Ranking within a class is calculated. Senior Signals are processed preferentially. Here you can spec ell signals with high or low energy, bandwidth etc. can be selected.

The order of the signals is intended to be a certain pre-selection meet if z. B. not all signals at the same time can be searched.

If PSK2 or PSK4 modulation is present, the precise estimation of the parameters carrier frequency, phase, baud rate and phase is of great advantage for a good starting point of PSK demodulation (cf. FIG. 2).

For additional estimation of carrier frequency and phase in the presence of a PSK2 or PSK4 modulation, the complex baseband signal x is additionally added to the intermediate signal or signals x ² and in a further development of the invention in an additional device (TR in FIG. 2) / or x ⁴ transformed. These additional intermediate signals are then subjected to a parameter estimation of a higher-order autorregressive model, preferably a sixth order. The estimated carrier frequency is then derived from the phase angle of the largest pole position of this autoregressive model and the phase of the estimated carrier frequency is calculated by calculating the line of a discrete Fourier transformation (DFT) at the location of the estimated carrier frequency.

The complex baseband signal is used to determine the baud rate x filtered more narrowly than estimated by the segmentation or when summarizing according to a Fast Fourier Transfor mation (FFT) the edge channels of the signal are not loaded considered. The amount square of the signal should be the Include baud rate over an autoregressive model higher order, preferably twelfth order can be.

The real autocovariance of the signal becomes order of the autoregressive model per data block and calculates the autoregressive model.

In order to additionally estimate the baud rate and phase when PSK2 or PSK4 modulation is present, in another development of the invention the complex baseband signal x is first subjected to filtering which limits the spectrum width of the signal and then into the further intermediate signal | x | ² transformed. This further intermediate signal is then subjected to a parameter estimation of an autoregressive model of higher order, preferably twelfth order, and the estimated baud rate is taken from the phase angle of the largest pole position of this autoregressive model. The phase of the estimated baud rate is then determined by calculating the line of a discrete Fourier transformation (DFT) at the location of the estimated baud rate.

The particular advantage of these two developments of The inventive method are that a fold calculation or a non-cooperative clarification of the Carrier frequency and phase or baud rate and phase possible is.

In Fig. 5 (1 or 2 ME in Fig.) Is a preferred embodiment of the arrangement shown OF INVENTION to the invention for performing the method according to the invention.

The complex baseband signal x to be examined is connected to an input terminal 8 . The input terminal 8 is connected on the one hand directly to the input of a first order linear predictor 20 and to the input of a second order linear predictor 21 and on the other hand to the inputs of three nonlinear signal transformers 10 (x ² ), 11 (x ⁴ ) and 12 (| x | ² ), which in turn are each connected on the output side to a third-order linear predictor 22-24 . The signal transformer 12 (| x | ²) is also connected to a further signal transformer 13 (m _{| x |} ²) and to the "+" input 141 of a differential element 14 , the "-" input 142 of which is connected to the output of the Signal transformer 13 (m _{| x |} ²) is connected. The output 140 of the differential element 14 is in turn connected to a fourth linear third-order predictor 25 and, via a further signal transformer 15 (x ² ), to a fifth linear third-order predictor 26 . The seven linear predictors 20-27 form the unit 2 of the autoregressive models and are in turn connected on the output side to the evaluation circuit 3 .

Depending on their order and on the type of their input signal, the predictors deliver the (preferably standardized) prediction errors α _d1 , α _d2 , α _s , α _f , α _c , α _m and α _t (cf. also FIGS. 3 and 4). The output of the evaluation circuit 3 is connected to the output terminal 9 , at which the control signal S (z. B. for the demodulator DM in Fig. 1 or 2) can be tapped.

This arrangement works in parallel, ie the complex baseband signal x is practically simultaneously at the time t _k either via the signal transformers 10-15 in the form of the intermediate signals generated by the transformations or even directly supplied to the individual linear predictors 20-26 and subjected to a parameter estimate there fen. The prediction errors α _d1 , α _d2 , α _s , α _f , α _c , α _m , α _t obtained therefrom are fed to the evaluation circuit 3 , which, for. B. after the type of ratio formation or multiple comparison with threshold values shown in FIG. 4, the present modulation type is determined on the basis of these values and outputs a corresponding control signal S to the output terminal 9 .

The next complex baseband signal can then be processed in the same way at time t _{k + 1} .

A major advantage of this embodiment is because of the parallelism of the operations in the Schnel implementation.

However, the embodiment of the modulation type detector (ME in FIG. 1 or 2) shown in FIG. 6 can also be used to carry out the method according to the invention.

In this embodiment, the individual work not parallel (practically at the same time), son executed serially (i.e. one after the other).

The arrangement consists of a signal transformer 1 which can be set in terms of its transformation behavior and which is connected to a linear predictor 2 which can be set in its order. The linear predictor 2 is via a first switching device 61 and a downstream first memory unit 5 with memory locations 51-57 for the prediction errors α _d1 , α _d2 , α _s , α _f , α _c , α _m and α _t with the evaluation circuit 3 connected.

The evaluation circuit 3 is connected on the output side to the output terminal 9 of the arrangement.

The input terminal 8 for the complex baseband signal x is connected via a second memory unit 4 and a second switching device 62 either via a detour line 63 directly or via the signal transformer 1 to the linear predictor 2 . First and second memory units 5 and 4 , first and second switching devices 61 and 62 and the signal transformer 1 , the linear predictor 2 and the evaluation circuit 3 are controlled by a common controller 7 .

The 5 concurrent Ar in the arrangement of FIG. Beitsvorgängen leads are temporally successively Runaway here. For this purpose, the complex baseband signal x is first stored in the second memory 4 at the time t _k .

To determine the prediction error α _d1, for example, the controller 7 connects the second memory 4 directly to the linear predictor 2 via the second switching device 62 and the detour line 63 , which has been set to the first order by the controller 7 . The result of the parameter estimation (α _d1 ) is stored by means of the control device 7 via the first switching device 61, for example in the memory location 51 of the first storage unit 5 .

Then z. B. the prediction error α _{d2 is} determined by setting the linear predictor 2 to the second order by means of the controller 7 and storing the result of the parameter estimation (α _d2 ) via the first switching device 61, for example in the memory location 52 of the first memory unit 5 .

The prediction errors α _s , α _f , α _c , α _m and α _t are then determined, for example, by connecting the second memory 4 via the second switching device 62 to the signal transformer 1, which is successively used as intermediate signals x ² , x ⁴ , | x | ², (| x | ²-m _{| x | ²} ) and (| x | ²-m _{| x | ²} ) ² he creates, one in the linear predictor 2 , which has been set to the third order Parameter estimates are subjected, the result of which α _s , α _f , α _c , α _n and α _t are stored via the first switching device 61 in the corresponding memory locations 53-57 of the first memory unit 5 .

The α values obtained in this way can then be fed into the evaluation circuit 3 and the type of modulation can be determined there on the basis of these values in the manner already described.

Of course, the chronological order of the determination of the individual α values can deviate from the sequence described above, or even be carried out only in part: i.e. instead of the order α _d1 , α _d2 , α _s , α _f , α _c , α _m , α _t are z. B. also the sequences α _s , α _d1 , α _f , α _c , α _t , α _d2 , α _m or α _d1 , α _f , α _c or α _t , α _m , α _s , α _f , α _d2, etc possible.

The next complex baseband signal x can then be processed in the same way at time t _{k + 1} .

A major advantage of this embodiment is in that it is relatively inexpensive, because in principle among other things only one signal transformer (instead of  several) and a linear predictor (instead of several ren) are needed.

In FIG. 5, the first Schalteinrich may advantageously processing of a demultiplexer 61 can be realized in form and the second switching device 62 in the form of a multiple xers.

Claims (17)

1.Procedure for modulation type detection, in which a complex baseband signal x modulated according to a certain, initially still unknown modulation type is transformed into a plurality of independent intermediate signals, which intermediate signals form the basis for determining the type of modulation in an evaluation circuit, characterized in that the individual Intermediate signals and / or the complex baseband signal x itself are each subjected to a parameter estimate of at least one autoregressive model and that from the estimated parameters and / or prediction errors derived from these parameters by comparing at least some of these parameters or prediction errors with one another and / or with predetermined threshold values the type of modulation is determined. 2. The method according to claim 1, characterized in that the individual intermediate signals each by a linear or nonlinear transformation from the complex base band signal x are generated and that preferably the more number of transformations nonlinear transformations preferably of lower order, preferably of he are first, second, third or fourth order. 3. The method according to any one of claims 1 or 2, characterized characterized that the individual transformations time Lich one after the other and / or preferably in parallel lel be carried out. 4. The method according to any one of the preceding claims, since characterized in that for the individual Zwischenensi gnale and / or the complex baseband signal x itself auto regressive models of different orders, preferably lower order, preferably first, second or third ter order can be used. 5. The method according to any one of the preceding claims characterized in that the autoregressive used Models based on Levinson or Schur algorithms work. 6. The method according to any one of the preceding claims, since characterized by that the prediction errors on Values 0 1 can be standardized. 7. The method according to any one of the preceding claims for the detection of digital types of modulation, in particular of ASK2, FSK2, MSK2, PSK2 and PSK4, and / or of harmonic signals, characterized in that the complex base band signal x directly the parameter estimation of an au toregressive model first Order and / or second order is subjected and / or first into the intermediate signals x ² and / or x ⁴ and / or | x | ² and / or (| x | ²-m _{| x | ²} ) and / or ( | x | ²-m _{| x | ²} ) ² is transformed and these intermediate signals are each subjected to a parameter estimation of an authoritative third-order gressive model. 8. The method according to claim 7, characterized in that the prediction errors derived from the individual parameter estimates and preferably normalized α _d1 , α _d2 , α _s , α _f , α _c , α _m , α _t in the evaluation circuit, at least in part, in pairs be set in relation and these individual ratios α _y / α _z (y, z ∈ {d1, d2, s, f, c, m, t}; y ≠ z) are individually compared with predetermined threshold values S _yz and that The present type of modulation is determined by multiple comparison of the differences [(α _y / α _z ) -S _yz ) resulting from these individual comparisons. 9. The method according to any one of claims 7 or 8 for additional estimation of carrier frequency and phase in the presence of a PSK2 or PSK4 modulation, characterized in that the complex baseband signal x additional Lich in the or the intermediate signals x ² and / or x ⁴ is transformed and these additional intermediate signals are subjected to a parameter estimate of a higher-order autoregressive model, preferably sixth order, and the estimated carrier frequency is derived from the phase angle of the largest pole position of this autoregressive model and the phase of the estimated carrier frequency is calculated the line of a discrete Fourier transformation is determined at the location of the estimated carrier frequency. 10. The method according to any one of claims 7 to 9 for additional estimation of baud rate and phase in the presence of PSK2 or PSK4 modulation, characterized in that the complex baseband signal x is first subjected to a filtering limiting the spectrum width of the signal and then into the further intermediate signal | x | ^{2 is} transformed that this additional intermediate signal is subjected to a parameter estimate of a higher-order autoregressive model, preferably twelfth order, and the estimated baud rate is taken from the phase angle of the largest pole position of this autoregressive model and the phase of the estimated baud rate is calculated the line of a discrete Fourier transformation is determined at the location of the estimated construction rate. 11. Arrangement for performing the method according to one of claims 1 to 10, characterized in

• - that at least one signal transformer ( 1 ; 10-15 ) is provided for transforming the complex baseband signal x,
• - That at least one linear predictor ( 2 ; 20-26 ) is provided for performing the parameter estimation of the at least one autoregressive model on the individual intermediate signals and / or the complex base band signal x itself;
• - That the evaluation circuit ( 3 ) for determining the type of modulation after the at least one linear predictor ( 2 ; 20-26 ) and the at least one signal transformer ( 1 ; 10-15 ) is connected upstream.

12. The arrangement according to claim 11, characterized in

• - That N linear predictors ( 20-26 ) and (NM) Si signal transformers ( 10-15 ) with 0 MN and N <1 are provided;
• - That the N linear predictors ( 20-26 ) on the output side are each directly connected to the evaluation circuit ( 3 );
• - That M of the N linear predictors ( 20 , 21 ) on the output side directly and the other (NM) of the N linear predictors ( 22-26 ) on the input side each via one of the (NM) signal transformers ( 10-15 ) with the input terminal ( 8 ) are connected for the complex baseband signal x.

13. Arrangement according to claim 12, characterized in

• - That a linear first-order predictor ( 20 ) and a linear second-order predictor ( 21 ) are each connected on the input side directly to the input terminal ( 8 ) for the complex baseband signal x;
• - That a first linear third-order predictor ( 22 ) on the input side via an x ² signal transformer ( 10 ), a second linear third-order predictor ( 23 ) on the input side via an x ⁴ signal transformer ( 11 ), a third linear third-order predictor ( 24 ) on the input side via a | x | ² signal transformer ( 12 ), a fourth linear predictor ( 25 ) on the input side via a (| x | ²-m _{| x | ²} ) signal transformer ( 12-14 ) and a fifth linear third-order predictor ( 26 ) on the input side via a ( | x | -m _{| x | ²} ) ² signal transformer ( 12-15 ) are connected to the input terminal ( 8 ) for the complex baseband signal x.

14. Arrangement according to claim 13, characterized in

• - That the (| x | ²-m _{| x | ²} ) signal transformer ( 12-14 ) from the | x | ² signal transformer ( 12 ), an m _{| x | ²} signal transformer ( 13 ) and a dif ferent element ( 14 ) is formed;
• - That the differential element ( 14 ) with its output ( 140 ) to the fourth linear third-order predictor ( 25 ) and with its "+" input ( 141 ) at the output of the x ² signal transformer ( 12 ) and with its "- "Input ( 142 ) at the output of the m _{| x | ²} signal transformer ( 13 ) is connected;
• - that the entrance of the m _{| x | ²} signal transformer ( 13 ) is connected to the output of the | x | ² signal transformer ( 12 ).

15. The arrangement according to claim 14, characterized in

• - That the (| x | ²-m _{| x | ²} ) ² signal transformer ( 12-15 ) contains an additional x ² signal transformer ( 15 )
• - That the additional x ² signal transformer ( 15 ) on the output side to the fifth linear third-order predictor ( 26 ) and on the input side to the output ( 140 ) of the differential element ( 14 ) of the (| x | ²-m _{| x | ²} ) signal transformer ( 12-14 ) is connected.

16. The arrangement according to claim 11, characterized in

• - That a signal transformer ( 1 ) and a linear predictor ( 2 ) are provided;
• - That the transformation behavior of the signal transformer ( 1 ) and the order of the linear predicator ( 2 ) are adjustable;
• - That the output of the linear predictor via a first switching device ( 61 ) and a first memory unit ( 5 ; 51-57 ) is connected to the evaluation circuit ( 3 );
• - That the input terminal ( 8 ) for the complex base band signal x via a second memory unit ( 4 ) and a second switching device ( 62 ) either with the input of the signal transformer ( 1 ) or via a bypass line ( 63 ) directly with the input of the linear predictor ( 2 ) is connectable;
• - That the signal transformer ( 1 ), the linear predictor ( 2 ), the evaluation circuit ( 3 ), the first and second memory unit ( 4 , 5 ) and the first and second switching device ( 61 , 62 ) by a common controller ( 7 ) are controlled.

17. The arrangement according to claim 16, characterized in that the first switching device ( 61 ) is realized in the form of a demul tiplexer and the second switching device ( 62 ) in the form of a multiplexer.